Existing deep learning (DL) based speech enhancement approaches are generally optimised to minimise the distance between clean and enhanced speech features. These often result in improved speech quality however they suffer from a lack of generalisation and may not deliver the required speech intelligibility in real noisy situations. In an attempt to address these challenges, researchers have explored intelligibility-oriented (I-O) loss functions and integration of audio-visual (AV) information for more robust speech enhancement (SE). In this paper, we introduce DL based I-O SE algorithms exploiting AV information, which is a novel and previously unexplored research direction. Specifically, we present a fully convolutional AV SE model that uses a modified short-time objective intelligibility (STOI) metric as a training cost function. To the best of our knowledge, this is the first work that exploits the integration of AV modalities with an I-O based loss function for SE. Comparative experimental results demonstrate that our proposed I-O AV SE framework outperforms audio-only (AO) and AV models trained with conventional distance-based loss functions, in terms of standard objective evaluation measures when dealing with unseen speakers and noises.
Automatic nuclei segmentation and classification plays a vital role in digital pathology. However, previous works are mostly built on data with limited diversity and small sizes, making the results questionable or misleading in actual downstream tasks. In this paper, we aim to build a reliable and robust method capable of dealing with data from the 'the clinical wild'. Specifically, we study and design a new method to simultaneously detect, segment, and classify nuclei from Haematoxylin and Eosin (H&E) stained histopathology data, and evaluate our approach using the recent largest dataset: PanNuke. We address the detection and classification of each nuclei as a novel semantic keypoint estimation problem to determine the center point of each nuclei. Next, the corresponding class-agnostic masks for nuclei center points are obtained using dynamic instance segmentation. By decoupling two simultaneous challenging tasks, our method can benefit from class-aware detection and class-agnostic segmentation, thus leading to a significant performance boost. We demonstrate the superior performance of our proposed approach for nuclei segmentation and classification across 19 different tissue types, delivering new benchmark results.
Object recognition from live video streams comes with numerous challenges such as the variation in illumination conditions and poses. Convolutional neural networks (CNNs) have been widely used to perform intelligent visual object recognition. Yet, CNNs still suffer from severe accuracy degradation, particularly on illumination-variant datasets. To address this problem, we propose a new CNN method based on orientation fusion for visual object recognition. The proposed cloud-based video analytics system pioneers the use of bi-dimensional empirical mode decomposition to split a video frame into intrinsic mode functions (IMFs). We further propose these IMFs to endure Reisz transform to produce monogenic object components, which are in turn used for the training of CNNs. Past works have demonstrated how the object orientation component may be used to pursue accuracy levels as high as 93\%. Herein we demonstrate how a feature-fusion strategy of the orientation components leads to further improving visual recognition accuracy to 97\%. We also assess the scalability of our method, looking at both the number and the size of the video streams under scrutiny. We carry out extensive experimentation on the publicly available Yale dataset, including also a self generated video datasets, finding significant improvements (both in accuracy and scale), in comparison to AlexNet, LeNet and SE-ResNeXt, which are the three most commonly used deep learning models for visual object recognition and classification.
Graph representation learning has become a ubiquitous component in many scenarios, ranging from social network analysis to energy forecasting in smart grids. In several applications, ensuring the fairness of the node (or graph) representations with respect to some protected attributes is crucial for their correct deployment. Yet, fairness in graph deep learning remains under-explored, with few solutions available. In particular, the tendency of similar nodes to cluster on several real-world graphs (i.e., homophily) can dramatically worsen the fairness of these procedures. In this paper, we propose a biased edge dropout algorithm (FairDrop) to counter-act homophily and improve fairness in graph representation learning. FairDrop can be plugged in easily on many existing algorithms, is efficient, adaptable, and can be combined with other fairness-inducing solutions. After describing the general algorithm, we demonstrate its application on two benchmark tasks, specifically, as a random walk model for producing node embeddings, and to a graph convolutional network for link prediction. We prove that the proposed algorithm can successfully improve the fairness of all models up to a small or negligible drop in accuracy, and compares favourably with existing state-of-the-art solutions. In an ablation study, we demonstrate that our algorithm can flexibly interpolate between biasing towards fairness and an unbiased edge dropout. Furthermore, to better evaluate the gains, we propose a new dyadic group definition to measure the bias of a link prediction task when paired with group-based fairness metrics. In particular, we extend the metric used to measure the bias in the node embeddings to take into account the graph structure.
Nonlinear models are known to provide excellent performance in real-world applications that often operate in non-ideal conditions. However, such applications often require online processing to be performed with limited computational resources. In this paper, we propose a new efficient nonlinear model for online applications. The proposed algorithm is based on the linear-in-the-parameters (LIP) nonlinear filters and their implementation as functional link adaptive filters (FLAFs). We focus here on a new effective and efficient approach for FLAFs based on frequency-domain adaptive filters. We introduce the class of frequency-domain functional link adaptive filters (FD-FLAFs) and propose a partitioned block approach for their implementation. We also investigate on the functional link expansions that provide the most significant benefits operating with limited resources in the frequency-domain. We present and compare FD-FLAFs with different expansions to identify the LIP nonlinear filters showing the best tradeoff between performance and computational complexity. Experimental results prove that the frequency domain LIP nonlinear filters can be considered as an efficient and effective solution for online applications, like the nonlinear acoustic echo cancellation.
Common horizontal bounding box (HBB)-based methods are not capable of accurately locating slender ship targets with arbitrary orientations in synthetic aperture radar (SAR) images. Therefore, in recent years, methods based on oriented bounding box (OBB) have gradually received attention from researchers. However, most of the recently proposed deep learning-based methods for OBB detection encounter the boundary discontinuity problem in angle or key point regression. In order to alleviate this problem, researchers propose to introduce some manually set parameters or extra network branches for distinguishing the boundary cases, which make training more diffcult and lead to performance degradation. In this paper, in order to solve the boundary discontinuity problem in OBB regression, we propose to detect SAR ships by learning polar encodings. The encoding scheme uses a group of vectors pointing from the center of the ship target to the boundary points to represent an OBB. The boundary discontinuity problem is avoided by training and inference directly according to the polar encodings. In addition, we propose an Intersect over Union (IOU) -weighted regression loss, which further guides the training of polar encodings through the IOU metric and improves the detection performance. Experiments on the Rotating SAR Ship Detection Dataset (RSSDD) show that the proposed method can achieve better detection performance over other comparison algorithms and other OBB encoding schemes, demonstrating the effectiveness of our method.
There has recently been growing interest in utilizing multimodal sensors to achieve robust lane line segmentation. In this paper, we introduce a novel multimodal fusion architecture from an information theory perspective, and demonstrate its practical utility using Light Detection and Ranging (LiDAR) camera fusion networks. In particular, we develop, for the first time, a multimodal fusion network as a joint coding model, where each single node, layer, and pipeline is represented as a channel. The forward propagation is thus equal to the information transmission in the channels. Then, we can qualitatively and quantitatively analyze the effect of different fusion approaches. We argue the optimal fusion architecture is related to the essential capacity and its allocation based on the source and channel. To test this multimodal fusion hypothesis, we progressively determine a series of multimodal models based on the proposed fusion methods and evaluate them on the KITTI and the A2D2 datasets. Our optimal fusion network achieves 85%+ lane line accuracy and 98.7%+ overall. The performance gap among the models will inform continuing future research into development of optimal fusion algorithms for the deep multimodal learning community.
Segmentation-based methods are widely used for scene text detection due to their superiority in describing arbitrary-shaped text instances. However, two major problems still exist: 1) current label generation techniques are mostly empirical and lack theoretical support, discouraging elaborate label design; 2) as a result, most methods rely heavily on text kernel segmentation which is unstable and requires deliberate tuning. To address these challenges, we propose a human cognition-inspired framework, termed, Conceptual Text Region Network (CTRNet). The framework utilizes Conceptual Text Regions (CTRs), which is a class of cognition-based tools inheriting good mathematical properties, allowing for sophisticated label design. Another component of CTRNet is an inference pipeline that, with the help of CTRs, completely omits the need for text kernel segmentation. Compared with previous segmentation-based methods, our approach is not only more interpretable but also more accurate. Experimental results show that CTRNet achieves state-of-the-art performance on benchmark CTW1500, Total-Text, MSRA-TD500, and ICDAR 2015 datasets, yielding performance gains of up to 2.0%. Notably, to the best of our knowledge, CTRNet is among the first detection models to achieve F-measures higher than 85.0% on all four of the benchmarks, with remarkable consistency and stability.
Most recent works on sentiment analysis have exploited the text modality. However, millions of hours of video recordings posted on social media platforms everyday hold vital unstructured information that can be exploited to more effectively gauge public perception. Multimodal sentiment analysis offers an innovative solution to computationally understand and harvest sentiments from videos by contextually exploiting audio, visual and textual cues. In this paper, we, firstly, present a first of its kind Persian multimodal dataset comprising more than 800 utterances, as a benchmark resource for researchers to evaluate multimodal sentiment analysis approaches in Persian language. Secondly, we present a novel context-aware multimodal sentiment analysis framework, that simultaneously exploits acoustic, visual and textual cues to more accurately determine the expressed sentiment. We employ both decision-level (late) and feature-level (early) fusion methods to integrate affective cross-modal information. Experimental results demonstrate that the contextual integration of multimodal features such as textual, acoustic and visual features deliver better performance (91.39%) compared to unimodal features (89.24%).
Persuasion aims at forming one's opinion and action via a series of persuasive messages containing persuader's strategies. Due to its potential application in persuasive dialogue systems, the task of persuasive strategy recognition has gained much attention lately. Previous methods on user intent recognition in dialogue systems adopt recurrent neural network (RNN) or convolutional neural network (CNN) to model context in conversational history, neglecting the tactic history and intra-speaker relation. In this paper, we demonstrate the limitations of a Transformer-based approach coupled with Conditional Random Field (CRF) for the task of persuasive strategy recognition. In this model, we leverage inter- and intra-speaker contextual semantic features, as well as label dependencies to improve the recognition. Despite extensive hyper-parameter optimizations, this architecture fails to outperform the baseline methods. We observe two negative results. Firstly, CRF cannot capture persuasive label dependencies, possibly as strategies in persuasive dialogues do not follow any strict grammar or rules as the cases in Named Entity Recognition (NER) or part-of-speech (POS) tagging. Secondly, the Transformer encoder trained from scratch is less capable of capturing sequential information in persuasive dialogues than Long Short-Term Memory (LSTM). We attribute this to the reason that the vanilla Transformer encoder does not efficiently consider relative position information of sequence elements.